Import libraries¶

In [1]:
#import Libraries
import plotly.express as ex
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from imblearn.over_sampling import SMOTE
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import confusion_matrix, accuracy_score, classification_report, recall_score
from sklearn.cluster import KMeans
from sklearn.decomposition import PCA
from sklearn.metrics import silhouette_score
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler

Read Data¶

In [2]:
# Read data 
bank = pd.read_csv(r"C:\Users\junej\Documents\Credit card churn prediction\BankChurners.csv", encoding='latin1')
In [3]:
bank.head()
uncoded_bank=bank
bank.head(3)
Out[3]:
CLIENTNUM Attrition_Flag Customer_Age Gender Dependent_count Education_Level Marital_Status Income_Category Card_Category Months_on_book ... Months_Inactive_12_mon Contacts_Count_12_mon Credit_Limit Total_Revolving_Bal Avg_Open_To_Buy Total_Amt_Chng_Q4_Q1 Total_Trans_Amt Total_Trans_Ct Total_Ct_Chng_Q4_Q1 Avg_Utilization_Ratio
0 768805383 Existing Customer 45 M 3 High School Married $60K - $80K Blue 39 ... 1 3 12691.0 777 11914.0 1.335 1144 42 1.625 0.061
1 818770008 Existing Customer 49 F 5 Graduate Single Less than $40K Blue 44 ... 1 2 8256.0 864 7392.0 1.541 1291 33 3.714 0.105
2 713982108 Existing Customer 51 M 3 Graduate Married $80K - $120K Blue 36 ... 1 0 3418.0 0 3418.0 2.594 1887 20 2.333 0.000

3 rows × 21 columns

In [4]:
bank.describe()
Out[4]:
CLIENTNUM Customer_Age Dependent_count Months_on_book Total_Relationship_Count Months_Inactive_12_mon Contacts_Count_12_mon Credit_Limit Total_Revolving_Bal Avg_Open_To_Buy Total_Amt_Chng_Q4_Q1 Total_Trans_Amt Total_Trans_Ct Total_Ct_Chng_Q4_Q1 Avg_Utilization_Ratio
count 1.012700e+04 10127.000000 10127.000000 10127.000000 10127.000000 10127.000000 10127.000000 10127.000000 10127.000000 10127.000000 10127.000000 10127.000000 10127.000000 10127.000000 10127.000000
mean 7.391776e+08 46.325960 2.346203 35.928409 3.812580 2.341167 2.455317 8631.953698 1162.814061 7469.139637 0.759941 4404.086304 64.858695 0.712222 0.274894
std 3.690378e+07 8.016814 1.298908 7.986416 1.554408 1.010622 1.106225 9088.776650 814.987335 9090.685324 0.219207 3397.129254 23.472570 0.238086 0.275691
min 7.080821e+08 26.000000 0.000000 13.000000 1.000000 0.000000 0.000000 1438.300000 0.000000 3.000000 0.000000 510.000000 10.000000 0.000000 0.000000
25% 7.130368e+08 41.000000 1.000000 31.000000 3.000000 2.000000 2.000000 2555.000000 359.000000 1324.500000 0.631000 2155.500000 45.000000 0.582000 0.023000
50% 7.179264e+08 46.000000 2.000000 36.000000 4.000000 2.000000 2.000000 4549.000000 1276.000000 3474.000000 0.736000 3899.000000 67.000000 0.702000 0.176000
75% 7.731435e+08 52.000000 3.000000 40.000000 5.000000 3.000000 3.000000 11067.500000 1784.000000 9859.000000 0.859000 4741.000000 81.000000 0.818000 0.503000
max 8.283431e+08 73.000000 5.000000 56.000000 6.000000 6.000000 6.000000 34516.000000 2517.000000 34516.000000 3.397000 18484.000000 139.000000 3.714000 0.999000
In [5]:
bank.isnull().sum()
Out[5]:
CLIENTNUM                   0
Attrition_Flag              0
Customer_Age                0
Gender                      0
Dependent_count             0
Education_Level             0
Marital_Status              0
Income_Category             0
Card_Category               0
Months_on_book              0
Total_Relationship_Count    0
Months_Inactive_12_mon      0
Contacts_Count_12_mon       0
Credit_Limit                0
Total_Revolving_Bal         0
Avg_Open_To_Buy             0
Total_Amt_Chng_Q4_Q1        0
Total_Trans_Amt             0
Total_Trans_Ct              0
Total_Ct_Chng_Q4_Q1         0
Avg_Utilization_Ratio       0
dtype: int64

Proportion of churn vs non churn¶

In [6]:
ex.pie(bank,names='Attrition_Flag',title='Proportion of churn vs not churn customers',hole=0.33)

As we can see, only 16% of the data samples represent churn customers; in the following steps, I will use SMOTE to upsample the churn samples to match them with the regular customer sample size

Data Preprocessing¶

In [7]:
#One Hot Encoding
bank.Attrition_Flag = bank.Attrition_Flag.replace({'Attrited Customer':1,'Existing Customer':0})
bank.Gender = bank.Gender.replace({'F':1,'M':0})
bank= pd.concat([bank,pd.get_dummies(bank['Education_Level']).drop(columns=['Unknown'])],axis=1)
bank = pd.concat([bank,pd.get_dummies(bank['Income_Category']).drop(columns=['Unknown'])],axis=1)
bank = pd.concat([bank,pd.get_dummies(bank['Marital_Status']).drop(columns=['Unknown'])],axis=1)
bank = pd.concat([bank,pd.get_dummies(bank['Card_Category']).drop(columns=['Platinum'])],axis=1)
bank.drop(columns = ['Education_Level','Income_Category','Marital_Status','Card_Category','CLIENTNUM'],inplace=True)

Here we one hot encode all the categorical features describing different statuses of a customer.

In [8]:
bank
Out[8]:
Attrition_Flag Customer_Age Gender Dependent_count Months_on_book Total_Relationship_Count Months_Inactive_12_mon Contacts_Count_12_mon Credit_Limit Total_Revolving_Bal ... $40K - $60K $60K - $80K $80K - $120K Less than $40K Divorced Married Single Blue Gold Silver
0 0 45 0 3 39 5 1 3 12691.0 777 ... 0 1 0 0 0 1 0 1 0 0
1 0 49 1 5 44 6 1 2 8256.0 864 ... 0 0 0 1 0 0 1 1 0 0
2 0 51 0 3 36 4 1 0 3418.0 0 ... 0 0 1 0 0 1 0 1 0 0
3 0 40 1 4 34 3 4 1 3313.0 2517 ... 0 0 0 1 0 0 0 1 0 0
4 0 40 0 3 21 5 1 0 4716.0 0 ... 0 1 0 0 0 1 0 1 0 0
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
10122 0 50 0 2 40 3 2 3 4003.0 1851 ... 1 0 0 0 0 0 1 1 0 0
10123 1 41 0 2 25 4 2 3 4277.0 2186 ... 1 0 0 0 1 0 0 1 0 0
10124 1 44 1 1 36 5 3 4 5409.0 0 ... 0 0 0 1 0 1 0 1 0 0
10125 1 30 0 2 36 4 3 3 5281.0 0 ... 1 0 0 0 0 0 0 1 0 0
10126 1 43 1 2 25 6 2 4 10388.0 1961 ... 0 0 0 1 0 1 0 0 0 1

10127 rows × 33 columns

Correlation matrix¶

In [9]:
correlation_matrix = bank.corr()
print(correlation_matrix)

                          Attrition_Flag  Customer_Age    Gender  \
Attrition_Flag                  1.000000      0.018203  0.037272   
Customer_Age                    0.018203      1.000000  0.017312   
Gender                          0.037272      0.017312  1.000000   
Dependent_count                 0.018991     -0.122254 -0.004563   
Months_on_book                  0.013687      0.788912  0.006728   
Total_Relationship_Count       -0.150005     -0.010931 -0.003157   
Months_Inactive_12_mon          0.152449      0.054361  0.011163   
Contacts_Count_12_mon           0.204491     -0.018452 -0.039987   
Credit_Limit                   -0.023873      0.002476 -0.420806   
Total_Revolving_Bal            -0.263053      0.014780 -0.029658   
Avg_Open_To_Buy                -0.000285      0.001151 -0.418059   
Total_Amt_Chng_Q4_Q1           -0.131063     -0.062042 -0.026712   
Total_Trans_Amt                -0.168598     -0.046446 -0.024890   
Total_Trans_Ct                 -0.371403     -0.067097  0.067454   
Total_Ct_Chng_Q4_Q1            -0.290054     -0.012143  0.005800   
Avg_Utilization_Ratio          -0.178410      0.007114  0.257851   
College                        -0.007840     -0.014788 -0.002610   
Doctorate                       0.029386      0.025199  0.017631   
Graduate                       -0.009046     -0.000203  0.006438   
High School                    -0.011730      0.001199 -0.018361   
Post-Graduate                   0.011127     -0.022081 -0.009001   
Uneducated                     -0.001444      0.005057  0.005174   
$120K +                         0.009584      0.044332 -0.294775   
$40K - $60K                    -0.011688     -0.013804  0.034718   
$60K - $80K                    -0.028221     -0.017869 -0.424892   
$80K - $120K                   -0.003459      0.005381 -0.448017   
Less than $40K                  0.022466     -0.002573  0.580016   
Divorced                        0.000850     -0.042614  0.004726   
Married                        -0.023735      0.047364 -0.011427   
Single                          0.019037     -0.011248  0.015756   
Blue                            0.003216      0.021409  0.085200   
Gold                            0.005973     -0.011901 -0.043453   
Silver                         -0.008467     -0.019425 -0.072699   

                          Dependent_count  Months_on_book  \
Attrition_Flag                   0.018991        0.013687   
Customer_Age                    -0.122254        0.788912   
Gender                          -0.004563        0.006728   
Dependent_count                  1.000000       -0.103062   
Months_on_book                  -0.103062        1.000000   
Total_Relationship_Count        -0.039076       -0.009203   
Months_Inactive_12_mon          -0.010768        0.074164   
Contacts_Count_12_mon           -0.040505       -0.010774   
Credit_Limit                     0.068065        0.007507   
Total_Revolving_Bal             -0.002688        0.008623   
Avg_Open_To_Buy                  0.068291        0.006732   
Total_Amt_Chng_Q4_Q1            -0.035439       -0.048959   
Total_Trans_Amt                  0.025046       -0.038591   
Total_Trans_Ct                   0.049912       -0.049819   
Total_Ct_Chng_Q4_Q1              0.011087       -0.014072   
Avg_Utilization_Ratio           -0.037135       -0.007541   
College                          0.003369       -0.010281   
Doctorate                       -0.003368        0.024114   
Graduate                         0.000671        0.003531   
High School                     -0.013127        0.002637   
Post-Graduate                    0.009459       -0.016703   
Uneducated                       0.002190        0.001099   
$120K +                          0.024831        0.033532   
$40K - $60K                     -0.013493       -0.003204   
$60K - $80K                      0.028975       -0.016635   
$80K - $120K                     0.047611        0.007720   
Less than $40K                  -0.046785       -0.004300   
Divorced                         0.006697       -0.027678   
Married                          0.014385        0.033194   
Single                          -0.040707       -0.005065   
Blue                            -0.027671        0.014934   
Gold                             0.027035       -0.005426   
Silver                           0.017322       -0.014357   

                          Total_Relationship_Count  Months_Inactive_12_mon  \
Attrition_Flag                           -0.150005                0.152449   
Customer_Age                             -0.010931                0.054361   
Gender                                   -0.003157                0.011163   
Dependent_count                          -0.039076               -0.010768   
Months_on_book                           -0.009203                0.074164   
Total_Relationship_Count                  1.000000               -0.003675   
Months_Inactive_12_mon                   -0.003675                1.000000   
Contacts_Count_12_mon                     0.055203                0.029493   
Credit_Limit                             -0.071386               -0.020394   
Total_Revolving_Bal                       0.013726               -0.042210   
Avg_Open_To_Buy                          -0.072601               -0.016605   
Total_Amt_Chng_Q4_Q1                      0.050119               -0.032247   
Total_Trans_Amt                          -0.347229               -0.036982   
Total_Trans_Ct                           -0.241891               -0.042787   
Total_Ct_Chng_Q4_Q1                       0.040831               -0.038989   
Avg_Utilization_Ratio                     0.067663               -0.007503   
College                                  -0.013582                0.004038   
Doctorate                                -0.009077                0.002432   
Graduate                                  0.005397                0.005885   
High School                              -0.001707               -0.005575   
Post-Graduate                             0.012050               -0.006240   
Uneducated                                0.008202                0.010127   
$120K +                                  -0.006090               -0.002661   
$40K - $60K                              -0.008747               -0.021437   
$60K - $80K                               0.005475               -0.004616   
$80K - $120K                              0.001185               -0.005910   
Less than $40K                            0.005109                0.018848   
Divorced                                  0.009276                0.001796   
Married                                   0.017001               -0.007065   
Single                                   -0.017329                0.008973   
Blue                                      0.085789                0.016565   
Gold                                     -0.055678               -0.003283   
Silver                                   -0.060566               -0.016034   

                          Contacts_Count_12_mon  Credit_Limit  \
Attrition_Flag                         0.204491     -0.023873   
Customer_Age                          -0.018452      0.002476   
Gender                                -0.039987     -0.420806   
Dependent_count                       -0.040505      0.068065   
Months_on_book                        -0.010774      0.007507   
Total_Relationship_Count               0.055203     -0.071386   
Months_Inactive_12_mon                 0.029493     -0.020394   
Contacts_Count_12_mon                  1.000000      0.020817   
Credit_Limit                           0.020817      1.000000   
Total_Revolving_Bal                   -0.053913      0.042493   
Avg_Open_To_Buy                        0.025646      0.995981   
Total_Amt_Chng_Q4_Q1                  -0.024445      0.012813   
Total_Trans_Amt                       -0.112774      0.171730   
Total_Trans_Ct                        -0.152213      0.075927   
Total_Ct_Chng_Q4_Q1                   -0.094997     -0.002020   
Avg_Utilization_Ratio                 -0.055471     -0.482965   
College                               -0.008996      0.001929   
Doctorate                             -0.001016     -0.005195   
Graduate                               0.002660     -0.004844   
High School                           -0.003927     -0.001432   
Post-Graduate                         -0.006878      0.005879   
Uneducated                             0.012596      0.012213   
$120K +                                0.020744      0.339211   
$40K - $60K                            0.002336     -0.161605   
$60K - $80K                            0.000167      0.093808   
$80K - $120K                           0.011473      0.333828   
Less than $40K                        -0.023255     -0.395233   
Divorced                              -0.008389      0.022578   
Married                                0.001598     -0.056713   
Single                                 0.007997      0.032309   
Blue                                   0.000929     -0.515644   
Gold                                  -0.000685      0.234331   
Silver                                -0.001059      0.441030   

                          Total_Revolving_Bal  ...  $40K - $60K  $60K - $80K  \
Attrition_Flag                      -0.263053  ...    -0.011688    -0.028221   
Customer_Age                         0.014780  ...    -0.013804    -0.017869   
Gender                              -0.029658  ...     0.034718    -0.424892   
Dependent_count                     -0.002688  ...    -0.013493     0.028975   
Months_on_book                       0.008623  ...    -0.003204    -0.016635   
Total_Relationship_Count             0.013726  ...    -0.008747     0.005475   
Months_Inactive_12_mon              -0.042210  ...    -0.021437    -0.004616   
Contacts_Count_12_mon               -0.053913  ...     0.002336     0.000167   
Credit_Limit                         0.042493  ...    -0.161605     0.093808   
Total_Revolving_Bal                  1.000000  ...     0.005220    -0.003847   
Avg_Open_To_Buy                     -0.047167  ...    -0.162039     0.094133   
Total_Amt_Chng_Q4_Q1                 0.058174  ...    -0.000867     0.022399   
Total_Trans_Amt                      0.064370  ...     0.000209     0.005502   
Total_Trans_Ct                       0.056060  ...     0.002801    -0.027517   
Total_Ct_Chng_Q4_Q1                  0.089861  ...    -0.002159     0.014402   
Avg_Utilization_Ratio                0.624022  ...     0.070035    -0.100068   
College                             -0.011058  ...     0.003405    -0.007854   
Doctorate                           -0.018208  ...    -0.012193    -0.004764   
Graduate                            -0.000356  ...     0.000061    -0.006836   
High School                          0.019276  ...    -0.000524     0.020287   
Post-Graduate                        0.007068  ...     0.023301     0.007234   
Uneducated                          -0.004446  ...    -0.010118    -0.008775   
$120K +                              0.020264  ...    -0.128862    -0.111479   
$40K - $60K                          0.005220  ...     1.000000    -0.185743   
$60K - $80K                         -0.003847  ...    -0.185743     1.000000   
$80K - $120K                         0.022776  ...    -0.195852    -0.169433   
Less than $40K                      -0.015974  ...    -0.341238    -0.295207   
Divorced                            -0.002368  ...     0.005728     0.004860   
Married                              0.039124  ...    -0.006464     0.006952   
Single                              -0.037794  ...     0.003745    -0.008723   
Blue                                -0.022681  ...     0.007328    -0.037802   
Gold                                 0.023974  ...    -0.013388     0.034771   
Silver                               0.012799  ...     0.001025     0.024076   

                          $80K - $120K  Less than $40K  Divorced   Married  \
Attrition_Flag               -0.003459        0.022466  0.000850 -0.023735   
Customer_Age                  0.005381       -0.002573 -0.042614  0.047364   
Gender                       -0.448017        0.580016  0.004726 -0.011427   
Dependent_count               0.047611       -0.046785  0.006697  0.014385   
Months_on_book                0.007720       -0.004300 -0.027678  0.033194   
Total_Relationship_Count      0.001185        0.005109  0.009276  0.017001   
Months_Inactive_12_mon       -0.005910        0.018848  0.001796 -0.007065   
Contacts_Count_12_mon         0.011473       -0.023255 -0.008389  0.001598   
Credit_Limit                  0.333828       -0.395233  0.022578 -0.056713   
Total_Revolving_Bal           0.022776       -0.015974 -0.002368  0.039124   
Avg_Open_To_Buy               0.331716       -0.393718  0.022786 -0.060209   
Total_Amt_Chng_Q4_Q1          0.012905       -0.025276 -0.005613  0.052388   
Total_Trans_Amt               0.009814       -0.008816  0.010392 -0.063030   
Total_Trans_Ct               -0.038939        0.040664  0.024863 -0.123911   
Total_Ct_Chng_Q4_Q1           0.002922       -0.001399 -0.011503  0.013639   
Avg_Utilization_Ratio        -0.173606        0.273511 -0.018385  0.049684   
College                       0.019690       -0.007724  0.014065 -0.001214   
Doctorate                    -0.015165       -0.000589  0.004920 -0.003584   
Graduate                      0.002308        0.017496 -0.004936  0.013413   
High School                   0.001987       -0.019091 -0.019568  0.008604   
Post-Graduate                 0.003490       -0.010762  0.004957  0.003768   
Uneducated                   -0.006529       -0.000514  0.027912 -0.018026   
$120K +                      -0.117547       -0.204804 -0.002483  0.013448   
$40K - $60K                  -0.195852       -0.341238  0.005728 -0.006464   
$60K - $80K                  -0.169433       -0.295207  0.004860  0.006952   
$80K - $120K                  1.000000       -0.311274 -0.010926  0.013568   
Less than $40K               -0.311274        1.000000 -0.007134 -0.008341   
Divorced                     -0.010926       -0.007134  1.000000 -0.262132   
Married                       0.013568       -0.008341 -0.262132  1.000000   
Single                       -0.020703        0.018028 -0.225502 -0.741185   
Blue                         -0.038508        0.069699 -0.001440  0.051688   
Gold                          0.008843       -0.032630 -0.012659 -0.023612   
Silver                        0.039775       -0.059205  0.008305 -0.044260   

                            Single      Blue      Gold    Silver  
Attrition_Flag            0.019037  0.003216  0.005973 -0.008467  
Customer_Age             -0.011248  0.021409 -0.011901 -0.019425  
Gender                    0.015756  0.085200 -0.043453 -0.072699  
Dependent_count          -0.040707 -0.027671  0.027035  0.017322  
Months_on_book           -0.005065  0.014934 -0.005426 -0.014357  
Total_Relationship_Count -0.017329  0.085789 -0.055678 -0.060566  
Months_Inactive_12_mon    0.008973  0.016565 -0.003283 -0.016034  
Contacts_Count_12_mon     0.007997  0.000929 -0.000685 -0.001059  
Credit_Limit              0.032309 -0.515644  0.234331  0.441030  
Total_Revolving_Bal      -0.037794 -0.022681  0.023974  0.012799  
Avg_Open_To_Buy           0.035691 -0.513502  0.232132  0.439790  
Total_Amt_Chng_Q4_Q1     -0.043323 -0.005710  0.006219  0.002621  
Total_Trans_Amt           0.044772 -0.194374  0.103986  0.154983  
Total_Trans_Ct            0.098586 -0.130755  0.076399  0.100971  
Total_Ct_Chng_Q4_Q1      -0.009774  0.005433 -0.002728 -0.003066  
Avg_Utilization_Ratio    -0.030493  0.214006 -0.085041 -0.190051  
College                  -0.005681 -0.005064  0.010505  0.003591  
Doctorate                 0.006284  0.003367 -0.000747 -0.005714  
Graduate                 -0.009162 -0.013194  0.000342  0.012746  
High School              -0.000899  0.012123  0.006841 -0.013394  
Post-Graduate            -0.010966 -0.008533 -0.003842  0.007342  
Uneducated                0.004022  0.006045 -0.015817  0.000621  
$120K +                  -0.007109 -0.049146  0.034771  0.033880  
$40K - $60K               0.003745  0.007328 -0.013388  0.001025  
$60K - $80K              -0.008723 -0.037802  0.034771  0.024076  
$80K - $120K             -0.020703 -0.038508  0.008843  0.039775  
Less than $40K            0.018028  0.069699 -0.032630 -0.059205  
Divorced                 -0.225502 -0.001440 -0.012659  0.008305  
Married                  -0.741185  0.051688 -0.023612 -0.044260  
Single                    1.000000 -0.040122  0.024426  0.031060  
Blue                     -0.040122  1.000000 -0.397782 -0.889816  
Gold                      0.024426 -0.397782  1.000000 -0.025920  
Silver                    0.031060 -0.889816 -0.025920  1.000000  

[33 rows x 33 columns]
In [10]:
plt.figure(figsize=(20,20))
sns.heatmap(correlation_matrix, annot=True, cmap='coolwarm', linewidths=0.5)
plt.title('Pearson Correlation Heatmap')
plt.show()

Logistic regression¶

In [11]:
# Separate features (X) and the target variable (y)
X = bank[bank.columns[1:]]  # Features
y = bank[bank.columns[0]]      # Target variable
x_train, x_test, y_train, y_test = train_test_split(X, bank['Attrition_Flag'], test_size = 0.2, random_state=1234)
In [12]:
# Initialize and train a logistic regression model
logistic_regression = LogisticRegression(max_iter=1000)  # You can adjust 'max_iter' as needed
logistic_regression.fit(x_train, y_train)

# Make predictions on the test set
y_pred = logistic_regression.predict(x_test)

# Calculate the confusion matrix
conf_matrix = confusion_matrix(y_test, y_pred)

# Calculate the recall score
recall = recall_score(y_test, y_pred)

# Print the confusion matrix
print("Confusion Matrix:")
print(conf_matrix)

# Evaluate the model's performance (accuracy, classification report)
accuracy = accuracy_score(y_test, y_pred)
classification_report_str = classification_report(y_test, y_pred)

# Print the results
print("Recall:", recall) 
print("Accuracy:", accuracy)
print("Classification Report:\n", classification_report_str)

Confusion Matrix:
[[1655   52]
 [ 152  167]]
Recall: 0.5235109717868338
Accuracy: 0.8993089832181639
Classification Report:
               precision    recall  f1-score   support

           0       0.92      0.97      0.94      1707
           1       0.76      0.52      0.62       319

    accuracy                           0.90      2026
   macro avg       0.84      0.75      0.78      2026
weighted avg       0.89      0.90      0.89      2026
In [13]:
plt.figure(figsize=(8, 6))
sns.heatmap(conf_matrix, annot=True, fmt='d', cmap='Blues', 
            xticklabels=['Predicted 0', 'Predicted 1'], 
            yticklabels=['Actual 0', 'Actual 1'])
plt.xlabel('Predicted')
plt.ylabel('Actual')
plt.title('Confusion Matrix')
plt.show()

Recall reveals the proportion of churns identified correctly by the total number of churns.In this model we have achieved a recall rate of 50%.

Smote and logistic regression¶

In [14]:
# Initialize and apply SMOTE to oversample the minority class
oversample = SMOTE()
X_resampled, y_resampled = oversample.fit_resample(X, y)

# Create a new DataFrame with the resampled data
resampled_df = pd.concat([pd.DataFrame(X_resampled, columns=X.columns), pd.Series(y_resampled, name='Churn')], axis=1)
In [15]:
correlation_matrix=resampled_df.corr()
plt.figure(figsize=(20,20))
sns.heatmap(correlation_matrix, annot=True, cmap='coolwarm', linewidths=0.5)
plt.title('Pearson Correlation Heatmap')
plt.show()
In [16]:
resampled_df
Out[16]:
Customer_Age Gender Dependent_count Months_on_book Total_Relationship_Count Months_Inactive_12_mon Contacts_Count_12_mon Credit_Limit Total_Revolving_Bal Avg_Open_To_Buy ... $60K - $80K $80K - $120K Less than $40K Divorced Married Single Blue Gold Silver Churn
0 45 0 3 39 5 1 3 12691.000000 777 11914.000000 ... 1 0 0 0 1 0 1 0 0 0
1 49 1 5 44 6 1 2 8256.000000 864 7392.000000 ... 0 0 1 0 0 1 1 0 0 0
2 51 0 3 36 4 1 0 3418.000000 0 3418.000000 ... 0 1 0 0 1 0 1 0 0 0
3 40 1 4 34 3 4 1 3313.000000 2517 796.000000 ... 0 0 1 0 0 0 1 0 0 0
4 40 0 3 21 5 1 0 4716.000000 0 4716.000000 ... 1 0 0 0 1 0 1 0 0 0
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
16995 38 1 3 33 4 2 1 4046.834158 0 4046.834158 ... 0 0 0 0 1 0 1 0 0 1
16996 56 1 1 51 3 4 2 3312.043325 881 2430.263996 ... 0 0 0 1 0 0 1 0 0 1
16997 43 0 2 32 4 3 1 1438.300000 644 793.666319 ... 0 0 0 0 0 0 1 0 0 1
16998 50 0 0 41 2 3 1 3427.622846 0 3427.622846 ... 0 0 0 0 0 1 1 0 0 1
16999 47 0 2 38 1 3 1 32037.279882 1181 30855.383056 ... 0 0 0 0 0 1 1 0 0 1

17000 rows × 33 columns

In [17]:
# Split the data into training and testing sets
X_train, X_test, y_train, y_test = train_test_split(X_resampled, y_resampled, test_size=0.2, random_state=42)

# Initialize the logistic regression model
logistic_reg = LogisticRegression()

# Fit the model on the training data
logistic_reg.fit(X_train, y_train)

# Predict the target values on the testing data
y_pred = logistic_reg.predict(X_test)

# Evaluate the model
classification_report(y_test, y_pred)

# Calculate the confusion matrix
conf_matrix = confusion_matrix(y_test, y_pred)

# Calculate the recall score
recall = recall_score(y_test, y_pred)

# Print the confusion matrix
print("Confusion Matrix:")
print(conf_matrix)

# Evaluate the model's performance (accuracy, classification report)
accuracy = accuracy_score(y_test, y_pred)
classification_report_str = classification_report(y_test, y_pred)

# Print the results
print("Accuracy:", accuracy)
print("Recall:",recall)
print("Classification Report:\n", classification_report_str)
plt.figure(figsize=(8, 6))
sns.heatmap(conf_matrix, annot=True, fmt='d', cmap='Blues', 
            xticklabels=['Predicted 0', 'Predicted 1'], 
            yticklabels=['Actual 0', 'Actual 1'])
plt.xlabel('Predicted')
plt.ylabel('Actual')
plt.title('Confusion Matrix')
plt.show()

C:\Users\junej\anaconda3\lib\site-packages\sklearn\linear_model\_logistic.py:814: ConvergenceWarning:

lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression


Confusion Matrix:
[[1313  363]
 [ 348 1376]]
Accuracy: 0.7908823529411765
Recall: 0.7981438515081206
Classification Report:
               precision    recall  f1-score   support

           0       0.79      0.78      0.79      1676
           1       0.79      0.80      0.79      1724

    accuracy                           0.79      3400
   macro avg       0.79      0.79      0.79      3400
weighted avg       0.79      0.79      0.79      3400

After using SMOTE (SMOTE selects k-nearest neighbors from the same class. It then generates synthetic examples by interpolating between the selected instance and its k-nearest neighbors. )we have achived a recall rate of 70%.

K Means Clsutering¶

For this analysis, I will be clustering credit card users based on their card activity, and then I will dig a little deeper to see what other attributes may be unique to each group.

In [18]:
cols = uncoded_bank.iloc[:, 9:].select_dtypes(['uint8', 'int64', 'float64']).columns
In [19]:
cols
Out[19]:
Index(['Months_on_book', 'Total_Relationship_Count', 'Months_Inactive_12_mon',
       'Contacts_Count_12_mon', 'Credit_Limit', 'Total_Revolving_Bal',
       'Avg_Open_To_Buy', 'Total_Amt_Chng_Q4_Q1', 'Total_Trans_Amt',
       'Total_Trans_Ct', 'Total_Ct_Chng_Q4_Q1', 'Avg_Utilization_Ratio'],
      dtype='object')
In [20]:
k_values = []
silhouette_scores=[]
for i in range(2,16):
    km = KMeans(n_clusters=i, init='k-means++', max_iter=500, n_init=10, random_state=1)
    data = StandardScaler().fit_transform(uncoded_bank[cols])
    km.fit(data)
    k_values.append(km.inertia_)
    silhouette_scores.append(silhouette_score(data, km.labels_))
# Print K values and corresponding Silhouette Scores
for k, score in zip(k_values, silhouette_scores):
    print(f"K={k}, Silhouette Score={score}")

K=103753.61777446116, Silhouette Score=0.20650462503448958
K=92137.3160829283, Silhouette Score=0.13826893294481118
K=82913.83055674178, Silhouette Score=0.1579085689998887
K=77909.45717622529, Silhouette Score=0.16047194400764273
K=74096.81242118758, Silhouette Score=0.13579879293950167
K=71151.9786183667, Silhouette Score=0.12264673133282393
K=68748.96808991095, Silhouette Score=0.12155307422987284
K=66469.25033778106, Silhouette Score=0.11760435326006441
K=64466.90538157148, Silhouette Score=0.11692241034233193
K=62778.65613839377, Silhouette Score=0.11643201103408396
K=61221.66970579319, Silhouette Score=0.11764345360079785
K=59967.26443416673, Silhouette Score=0.11682344276252803
K=58612.71350106147, Silhouette Score=0.11479535312915867
K=57686.826737998366, Silhouette Score=0.11367270215947615
In [21]:
fig = plt.figure(figsize=(10,8))
ax = sns.lineplot(x=range(2,16), y=k_values)
ax.set_title('Elbow Method')
ax.title.set_size(20)
plt.xlabel('K Clusters')
plt.ylabel('Within-cluster Sums of Squares')
plt.show()

It looks like the optimal number of clusters is somewhere between 4 and 6. After trying each number, 6 clusters returned the highest Silhouette Score, so I decided to proceed with 6 clusters. The Silhoutte score is a measure of how well defined clusters are, with scores near 1 indicating well-defined clusters, and scores near 0 indicating overlapping clusters. Now let's create a pipeline that will first process our data and then create a KMeans model.

For preprocessing, we will first normalize our data, and then perform Principal Component Analysis.

Principal Component Analysis, or PCA, is a method that compresses the information found in the explanatory variables into a reduced number of new variables. This reduces the noise and dimensionality of the data while preserving most of the original information.

In [22]:
preprocessor = Pipeline(
[
 ("scaler", StandardScaler()),
 ("pca", PCA(n_components=2, random_state=1))
]
)

clusterer = Pipeline(
     [
       (
              "kmeans",
               KMeans(
                   n_clusters=6,
                   init="k-means++",
                   n_init=50,
                   max_iter=1000,
                   random_state=1
                      )
       )
      ]
   )

pipe = Pipeline(
        [
            ("preprocessor", preprocessor),
            ("clusterer", clusterer)
        ]
    )
In [23]:
# Fit the pipeline to your data
# Assuming you have a dataset 'X' to fit the pipeline on
pipe.fit(uncoded_bank[cols])

# Access the explained variance ratios
explained_variance_ratios = pipe.named_steps['preprocessor'].named_steps['pca'].explained_variance_ratio_

# Print the variance explained by both components
print('Variance Explained by Component 1:', explained_variance_ratios[0])
print('Variance Explained by Component 2:', explained_variance_ratios[1])

Variance Explained by Component 1: 0.21305681674068852
Variance Explained by Component 2: 0.16749519637408963

PCA1 and PCA2 explain 38% of variance, now let's add the cluster labels to our data and see how well our clusters are defined.PCA can help in improving the separation of clusters by emphasizing the directions of maximum variance. This can potentially lead to better-defined clusters.

In [37]:
preprocessed_data = pipe["preprocessor"].transform(uncoded_bank[cols])

predicted_labels = pipe["clusterer"]["kmeans"].labels_
In [38]:
pcadf = pd.DataFrame(
        pipe["preprocessor"].transform(uncoded_bank[cols]),
        columns=["component_1", "component_2"]
    )
   
pcadf["predicted_cluster"] = pipe["clusterer"]["kmeans"].labels_

plt.figure(figsize=(8, 8))
   
sp = sns.scatterplot(
        x="component_1",
        y="component_2",
        s=50,
        data=pcadf,
        hue="predicted_cluster",
        palette='tab10'
    )
   
sp.set_title("Clustering results from credit card use")
ax.title.set_size(20)
plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.0)
   
plt.show()

As we saw from the Silhouette score, there is clearly some overlap between clusters, but overall there appears to be several distinct clusters in the data.

In [39]:
uncoded_bank['cluster'] = pipe['clusterer']['kmeans'].labels_
uncoded_bank.head()
Out[39]:
CLIENTNUM Attrition_Flag Customer_Age Gender Dependent_count Education_Level Marital_Status Income_Category Card_Category Months_on_book ... Contacts_Count_12_mon Credit_Limit Total_Revolving_Bal Avg_Open_To_Buy Total_Amt_Chng_Q4_Q1 Total_Trans_Amt Total_Trans_Ct Total_Ct_Chng_Q4_Q1 Avg_Utilization_Ratio cluster
0 768805383 0 45 0 3 High School Married $60K - $80K Blue 39 ... 3 12691.0 777 11914.0 1.335 1144 42 1.625 0.061 3
1 818770008 0 49 1 5 Graduate Single Less than $40K Blue 44 ... 2 8256.0 864 7392.0 1.541 1291 33 3.714 0.105 1
2 713982108 0 51 0 3 Graduate Married $80K - $120K Blue 36 ... 0 3418.0 0 3418.0 2.594 1887 20 2.333 0.000 1
3 769911858 0 40 1 4 High School Unknown Less than $40K Blue 34 ... 1 3313.0 2517 796.0 1.405 1171 20 2.333 0.760 1
4 709106358 0 40 0 3 Uneducated Married $60K - $80K Blue 21 ... 0 4716.0 0 4716.0 2.175 816 28 2.500 0.000 1

5 rows × 22 columns

In [27]:
fig, axs = plt.subplots(figsize=(12,8))
ax = uncoded_bank['cluster'].value_counts().plot.pie(autopct='%1.0f%%')
ax.set_ylabel('')
ax.set_title('Cluster Composition')
ax.title.set_size(20)
plt.show()

Difference between various clusters¶

Turnover¶

In [28]:
fig, axs = plt.subplots(ncols=2,nrows=3, figsize=(20,14))

uncoded_bank[uncoded_bank['cluster'] == 0]['Attrition_Flag'].value_counts().sort_index().plot.barh(ax=axs[0,0], title='Group 1')
uncoded_bank[uncoded_bank['cluster'] == 1]['Attrition_Flag'].value_counts().sort_index().plot.barh(ax=axs[0,1], title='Group 2')
uncoded_bank[uncoded_bank['cluster'] == 2]['Attrition_Flag'].value_counts().sort_index().plot.barh(ax=axs[1,0], title='Group 3')
uncoded_bank[uncoded_bank['cluster'] == 3]['Attrition_Flag'].value_counts().sort_index().plot.barh(ax=axs[1,1], title='Group 4')
uncoded_bank[uncoded_bank['cluster'] == 4]['Attrition_Flag'].value_counts().sort_index().plot.barh(ax=axs[2,0], title='Group 5')
uncoded_bank[uncoded_bank['cluster'] == 5]['Attrition_Flag'].value_counts().sort_index().plot.barh(ax=axs[2,1], title='Group 6')
plt.show()

It looks like Groups 2 and 4 have a significantly higher rate of customer attrition, while Group 1 has a significantly lower rate of attrition.

Account activity¶

In [29]:
scaler = StandardScaler()
scaled_cc = pd.DataFrame(scaler.fit_transform(uncoded_bank[cols]), columns=cols)
scaled_cc['cluster'] = pipe['clusterer']['kmeans'].labels_
scaled_cc.head()
Out[29]:
Months_on_book Total_Relationship_Count Months_Inactive_12_mon Contacts_Count_12_mon Credit_Limit Total_Revolving_Bal Avg_Open_To_Buy Total_Amt_Chng_Q4_Q1 Total_Trans_Amt Total_Trans_Ct Total_Ct_Chng_Q4_Q1 Avg_Utilization_Ratio cluster
0 0.384621 0.763943 -1.327136 0.492404 0.446622 -0.473422 0.488971 2.623494 -0.959707 -0.973895 3.834003 -0.775882 3
1 1.010715 1.407306 -1.327136 -0.411616 -0.041367 -0.366667 -0.008486 3.563293 -0.916433 -1.357340 12.608573 -0.616276 1
2 0.008965 0.120579 -1.327136 -2.219655 -0.573698 -1.426858 -0.445658 8.367214 -0.740982 -1.911206 6.807864 -0.997155 1
3 -0.241473 -0.522785 1.641478 -1.315636 -0.585251 1.661686 -0.734100 2.942843 -0.951758 -1.911206 6.807864 1.759686 1
4 -1.869317 0.763943 -1.327136 -2.219655 -0.430877 -1.426858 -0.302868 6.455682 -1.056263 -1.570365 7.509325 -0.997155 1
In [30]:
ix, axs = plt.subplots(ncols=2,nrows=3, figsize=(20,32))

scaled_cc[scaled_cc['cluster'] == 0][cols].mean().plot.barh(ax=axs[0,0], xlim=(-1.5, 3), figsize=(20,20), sharey=True, title='Group 1')
scaled_cc[scaled_cc['cluster'] == 1][cols].mean().plot.barh(ax=axs[0,1], xlim=(-1.5, 3), figsize=(20,20), sharey=True, title='Group 2')
scaled_cc[scaled_cc['cluster'] == 2][cols].mean().plot.barh(ax=axs[1,0], xlim=(-1.5, 3), figsize=(20,20), sharey=True, title='Group 3')
scaled_cc[scaled_cc['cluster'] == 3][cols].mean().plot.barh(ax=axs[1,1], xlim=(-1.5, 3), figsize=(20,20), sharey=True, title='Group 4')
scaled_cc[scaled_cc['cluster'] == 4][cols].mean().plot.barh(ax=axs[2,0], xlim=(-1.5, 3), figsize=(20,20), sharey=True, title='Group 5')
scaled_cc[scaled_cc['cluster'] == 5][cols].mean().plot.barh(ax=axs[2,1], xlim=(-1.5, 3), figsize=(20,20), sharey=True, title='Group 6')
plt.show()

Looking at the charts above, we can see the groups have the following characteristics:

Group 1: High transaction count and amount, higher balance, low relationship count, lower contact count

Group 2: Low transaction count and amount, lower count and amount change from Q4 to Q1, high contact count, higher relationship count, higher months on book, higher months inactive, lower balance

Group 3: High utilization ratio, higher count and amount change from Q4 to Q1, low avg open to buy, higher balance, low credit limit

Group 4: Low transaction count and amount, low utilization ratio, lower count and amount change from Q4 to Q1, high contact count, higher relationship count, lower balance

Group 5: Lower utilization ratio, higher transaction count, higher transaction count change from Q1 to Q4, low balance, low credit limit

Group 6: High transaction count and amount, high avg open to buy, high credit limit, low relationship count, low utilization ratio

Based on the results above, I decided to give the following names to each of the groups:

Group 1: Higher balance heavy users

Group 2: Long-time very light users

Group 3: Low limit, high balance light users

Group 4: High limit, low balance light users

Group 5: Low limit, low balance light users

Group 6: High limit, low balance heavy users

Age¶

In [31]:
fig, axs = plt.subplots(ncols=2,nrows=3, figsize=(12,14))

sns.histplot(uncoded_bank[uncoded_bank['cluster'] == 0]['Customer_Age'], ax=axs[0,0], binrange=(20,70), binwidth=5).set(title='Higher balance heavy users', ylabel='', xlabel='')
sns.histplot(uncoded_bank[uncoded_bank['cluster'] == 1]['Customer_Age'], ax=axs[0,1], binrange=(20,70), binwidth=5).set(title='Long-time very light users', ylabel='', xlabel='')
sns.histplot(uncoded_bank[uncoded_bank['cluster'] == 2]['Customer_Age'], ax=axs[1,0], binrange=(20,70), binwidth=5).set(title='Low limit, high balance light users', ylabel='', xlabel='')
sns.histplot(uncoded_bank[uncoded_bank['cluster'] == 3]['Customer_Age'], ax=axs[1,1], binrange=(20,70), binwidth=5).set(title='High limit, low balance light users', ylabel='', xlabel='')
sns.histplot(uncoded_bank[uncoded_bank['cluster'] == 4]['Customer_Age'], ax=axs[2,0], binrange=(20,70), binwidth=5).set(title='Low limit, low balance light users', ylabel='', xlabel='')
sns.histplot(uncoded_bank[uncoded_bank['cluster'] == 5]['Customer_Age'], ax=axs[2,1], binrange=(20,70), binwidth=5).set(title='High limit, low balance heavy users', ylabel='', xlabel='')
plt.show()

It looks like the majority of users are in their 40's and 50's for all groups. Long-time light users have a higher number of users in their 60's compared to other groups. Long-time light users and Low-limit high balance light users seem to have a higher number of users in their 30's.

Gender¶

In [32]:
fix, axs = plt.subplots(ncols=2,nrows=3, figsize=(12,14))

uncoded_bank[uncoded_bank['cluster'] == 0]['Gender'].value_counts().sort_index(ascending=False).plot.pie(ax=axs[0,0], ylabel='', title='Higher balance heavy users', autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 1]['Gender'].value_counts().sort_index(ascending=False).plot.pie(ax=axs[0,1], ylabel='', title='Long-time very light users', autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 2]['Gender'].value_counts().sort_index(ascending=False).plot.pie(ax=axs[1,0], ylabel='', title='Low limit, high balance light users', autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 3]['Gender'].value_counts().sort_index(ascending=False).plot.pie(ax=axs[1,1], ylabel='', title='High limit, low balance light users', autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 4]['Gender'].value_counts().sort_index(ascending=False).plot.pie(ax=axs[2,0], ylabel='', title='Low limit, low balance light users', autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 5]['Gender'].value_counts().sort_index(ascending=False).plot.pie(ax=axs[2,1], ylabel='', title='High limit, low balance heavy users', autopct='%.1f%%')
plt.show()
In [33]:
fix, axs = plt.subplots(ncols=2,nrows=3, figsize=(12,14))

uncoded_bank[uncoded_bank['cluster'] == 0]['Dependent_count'].value_counts().sort_index().plot.bar(ax=axs[0,0], title='Higher balance heavy users')
uncoded_bank[uncoded_bank['cluster'] == 1]['Dependent_count'].value_counts().sort_index().plot.bar(ax=axs[0,1], title='Long-time very light users')
uncoded_bank[uncoded_bank['cluster'] == 2]['Dependent_count'].value_counts().sort_index().plot.bar(ax=axs[1,0], title='Low limit, high balance light users')
uncoded_bank[uncoded_bank['cluster'] == 3]['Dependent_count'].value_counts().sort_index().plot.bar(ax=axs[1,1], title='High limit, low balance light users')
uncoded_bank[uncoded_bank['cluster'] == 4]['Dependent_count'].value_counts().sort_index().plot.bar(ax=axs[2,0], title='Low limit, low balance light users')
uncoded_bank[uncoded_bank['cluster'] == 5]['Dependent_count'].value_counts().sort_index().plot.bar(ax=axs[2,1], title='High limit, low balance heavy users')
plt.show()
In [34]:
fix, axs = plt.subplots(ncols=2,nrows=3, figsize=(12,14))

uncoded_bank[uncoded_bank['cluster'] == 0]['Education_Level'].value_counts().sort_index().plot.pie(ax=axs[0,0], ylabel='', title='Higher balance heavy users', colors=['r','b','g','y','orange', 'gray','pink'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 1]['Education_Level'].value_counts().sort_index().plot.pie(ax=axs[0,1], ylabel='', title='Long-time very light users', colors=['r','b','g','y','orange', 'gray','pink'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 2]['Education_Level'].value_counts().sort_index().plot.pie(ax=axs[1,0], ylabel='', title='Low limit, high balance light users', colors=['r','b','g','y','orange', 'gray','pink'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 3]['Education_Level'].value_counts().sort_index().plot.pie(ax=axs[1,1], ylabel='', title='High limit, low balance light users', colors=['r','b','g','y','orange', 'gray','pink'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 4]['Education_Level'].value_counts().sort_index().plot.pie(ax=axs[2,0], ylabel='', title='Low limit, low balance light users', colors=['r','b','g','y','orange', 'gray','pink'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 5]['Education_Level'].value_counts().sort_index().plot.pie(ax=axs[2,1], ylabel='', title='High limit, low balance heavy users', colors=['r','b','g','y','orange', 'gray','pink'], autopct='%.1f%%')
plt.show()

There doesn't seem to be any significant differences in education between groups.

Marital Status¶

In [35]:
fix, axs = plt.subplots(ncols=2,nrows=3, figsize=(12,14))

uncoded_bank[uncoded_bank['cluster'] == 0]['Marital_Status'].value_counts().sort_index().plot(kind='pie', ax=axs[0,0], ylabel='', title='Higher balance heavy users', colors=['r','b','g','y'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 1]['Marital_Status'].value_counts().sort_index().plot(kind='pie', ax=axs[0,1], ylabel='', title='Long-time very light users', colors=['r','b','g','y'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 2]['Marital_Status'].value_counts().sort_index().plot(kind='pie', ax=axs[1,0], ylabel='', title='Low limit, high balance light users', colors=['r','b','g','y'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 3]['Marital_Status'].value_counts().sort_index().plot(kind='pie', ax=axs[1,1], ylabel='', title='High limit, low balance light users', colors=['r','b','g','y'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 4]['Marital_Status'].value_counts().sort_index().plot(kind='pie', ax=axs[2,0], ylabel='', title='Low limit, low balance light users', colors=['r','b','g','y'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 5]['Marital_Status'].value_counts().sort_index().plot(kind='pie', ax=axs[2,1], ylabel='', title='High limit, low balance heavy users', colors=['r','b','g','y'], autopct='%.1f%%')
plt.show()

It looks like card limit and income are significantly correlated, with at least 60% of users in the high limit groups making at least 80K, and at least 50% of users in the low limit and high balance groups making less than 60K.

Card Type¶

In [36]:
fix, axs = plt.subplots(ncols=2,nrows=3, figsize=(12,14))

uncoded_bank[uncoded_bank['cluster'] == 0]['Card_Category'].value_counts().sort_index().plot(kind='pie', ax=axs[0,0], ylabel='', title='Higher balance heavy users', colors=['r','b','g','y'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 1]['Card_Category'].value_counts().sort_index().plot(kind='pie', ax=axs[0,1], ylabel='', title='Long-time very light users', colors=['r','b','g','y'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 2]['Card_Category'].value_counts().sort_index().plot(kind='pie', ax=axs[1,0], ylabel='', title='Low limit, high balance light users', colors=['r','b','g','y'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 3]['Card_Category'].value_counts().sort_index().plot(kind='pie', ax=axs[1,1], ylabel='', title='High limit, low balance light users', colors=['r','b','g','y'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 4]['Card_Category'].value_counts().sort_index().plot(kind='pie', ax=axs[2,0], ylabel='', title='Low limit, low balance light users', colors=['r','b','g','y'], autopct='%.1f%%')
uncoded_bank[uncoded_bank['cluster'] == 5]['Card_Category'].value_counts().sort_index().plot(kind='pie', ax=axs[2,1], ylabel='', title='High limit, low balance heavy users', colors=['r','b','g','y'], autopct='%.1f%%')
plt.show()

As expected, Silver, Gold, and Platinum card holders mostly fall into the high limit groups.

Conclusion Based on the our model, we clustered the credit card users into the following groups:

Group 1: Higher balance heavy users

Group 2: Long-time very light users

Group 3: Low limit, high balance light users

Group 4: High limit, low balance light users

Group 5: Low limit, low balance light users

Group 6: High limit, low balance heavy users

Looking at the demographic data of each group, we also found the following:

Long-time light users are the most likely to turnover.

Most users are in their 40's and 50's.

High limit users are mostly male.

The ratio of married users is higher among Long-time light users.

High limit users have significantly higher income than other groups, and are more likely to carry premium cards.

Through this analysis we have gained a good understanding of the groups that exist within the customers, as well as some of the attributes of each group.

In [ ]: